Approximating prediction uncertainty for random forest regression models

  • Authors: Coulston, John W.; Blinn, Christine E.; Thomas, Valerie A.; Wynne, Randolph H.
  • Publication Year: 2016
  • Publication Series: Scientific Journal (JRNL)
  • Source: Photogrammetric Engineering & Remote Sensing
  • DOI: 10.14358/PERS.82.3.189

Abstract

Machine learning approaches such as random forest have increased for the spatial modeling and mapping of continuous variables. Random forest is a non-parametric ensemble approach, and unlike traditional regression approaches there is no direct quantification of prediction error. Understanding prediction uncertainty is important when using model-based continuous maps as inputs to other modeling applications such as fire modeling. Here we use a Monte Carlo approach to quantify prediction uncertainty for random forest regression models. We test the approach by simulating maps of dependent and independent variables with known characteristics and comparing actual errors with prediction errors. Our approach produced conservative prediction intervals across most of the range of predicted values. However, because the Monte Carlo approach was data driven, prediction intervals were either too wide or too narrow in sparse parts of the prediction distribution. Overall, our approach provides reasonable estimates of prediction uncertainty for random forest regression models.

  • Citation: Coulston, John W.; Blinn, Christine E.; Thomas, Valerie A.; Wynne, Randolph H. 2016. Approximating prediction uncertainty for random forest regression models. Photogrammetric Engineering & Remote Sensing, Vol. 82(3): 189-197. 9 p.  10.14358/PERS.82.3.189
  • Posted Date: March 2, 2016
  • Modified Date: March 7, 2016
  • Print Publications Are No Longer Available

    In an ongoing effort to be fiscally responsible, the Southern Research Station (SRS) will no longer produce and distribute hard copies of our publications. Many SRS publications are available at cost via the Government Printing Office (GPO). Electronic versions of publications may be downloaded, printed, and distributed.

    Publication Notes

    • This article was written and prepared by U.S. Government employees on official time, and is therefore in the public domain.
    • Our online publications are scanned and captured using Adobe Acrobat. During the capture process some typographical errors may occur. Please contact the SRS webmaster if you notice any errors which make this publication unusable.
    • To view this article, download the latest version of Adobe Acrobat Reader.